[
https://issues.apache.org/jira/browse/FLINK-27594?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17536266#comment-17536266
]
Thomas Weise commented on FLINK-27594:
--------------------------------------
[~gyfora] are you saying that when no HA metadata is available and the upgrade
mode is LAST_STATE then the operator should keep the deployment in error state?
I think that would be correct. When the upgrade mode is SAVEPOINT, then it can
go back to that savepoint?
I also think that with LAST_STATE we should pick either last checkpoint or
savepoint, whichever is more recent.
> Only recover JM deployment if HA metadata available
> ---------------------------------------------------
>
> Key: FLINK-27594
> URL: https://issues.apache.org/jira/browse/FLINK-27594
> Project: Flink
> Issue Type: Bug
> Components: Kubernetes Operator
> Reporter: Gyula Fora
> Assignee: Gyula Fora
> Priority: Blocker
> Fix For: kubernetes-operator-1.0.0
>
>
> This ticket is related to https://issues.apache.org/jira/browse/FLINK-27572
> The deployment recovery logic for list jobmanager deployments simply performs
> a restoreFromLasteSavepoint operation currently.
> This is incorrect in cases where the HA metadata is not available as it might
> lead to accidentally restoring from an older state.
> We should verify that HA metadata is present and simply perform a
> deployOperation. Once we have this we can actually make the recovery default
> true for all versions.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
